Over the past few years, we’ve seen AI disrupt just about every industry out there. But some fields are particularly primed for an AI-driven revolution. This is the case for the life sciences, where there are vast amounts of data to manage and pressing problems to solve. It is little wonder, then, that AI is already transforming the field, from drug development to patient care to workflows.
 
We ask two contributors, Dr Silvia Quarteroni, Head of Innovation at the Swiss Data Science Center (SDSC), and Dr Nora Toussaint, Head of Biomedical Data Science at SDSC, to talk us through five ways AI is supercharging clinical innovation.

1. Speeding up drug development

We start by talking about the development of new drugs. When you design a drug, you must find a target. This is normally a molecule or structure in the body you aim to interact with to produce an effect. In the past, identifying targets involved lengthy and expensive wet lab work.

Now, Nora explains, AI is making the process easier and quicker: ‘For example, Insilico Medicine used an AI-driven pipeline to develop a drug candidate for idiopathic pulmonary fibrosis, an aggressive lung disease. In about 18 months, they went from an AI-identified target to a preclinical candidate, and they reached first-in-human testing in under 30 months. This is incredibly fast for pharma. Today, many early stages can be optimised in silico – on a computer – long before a pipette ever touches a sample or a human subject is involved.’

Nora is enthusiastic about the new technology that has led to this progress: ‘Over the last few years, AI has reshaped drug development and screening.’ AI model AlphaFold 3, for example, predicts structures and interactions across a broad range of biological molecules, from proteins to other molecules like DNA and RNA, making it much easier to find targets.

Silvia adds: ‘Entire research labs that used to carry out drug discovery and investigate molecules and proteins in manual and cumbersome ways have completely changed their approach. AlphaFold 3 has made the process faster and more efficient.’

2. Repurposing drugs to tackle rare diseases

Beyond developing new drugs, AI can be used to repurpose existing drugs for other conditions. This is especially useful for rare diseases, for which there are limited patients and so less usable data for developing new therapies.

Silvia elaborates on this idea: ‘What is really interesting about these deep neural network-based learning techniques is that they have the potential to port a model that has been developed on a large number of patients into much narrower domains. From these huge data sets, we can personalise treatment for one patient. They unlock possibilities when it comes to solving conditions for individual patients.’ This will help in the development of personalised medicine, which over time will gradually become cheaper thanks to new technology.

3. Unblocking bottlenecks in clinical development

In the life sciences, AI is also impacting the way workflows are organised – for instance, by alleviating bottlenecks in clinical development.

As Nora explains: ‘When you start a clinical trial, there are countless documents to write. You need a protocol that fulfils a long list of stringent requirements, which can take months to put together. LLMs – large language models – are a tremendous help. There are tools out there that will follow your guidelines and produce a document that ticks all the boxes.’ Although such documents still need to be thoroughly checked by a human being, this kind of AI-assisted workflow can be a huge time-saver.

Nora goes on to talk about how tricky it can be to recruit patients for clinical trials. ‘When you design the trial, you decide on the criteria and number of participants you’ll need, but it’s not always easy to find these participants. Again, this is something AI can help with, as it can efficiently go through health records to help identify suitable participants.’

Silvia explains that AI can go one step further through its predictive capabilities: ‘We’ve put together machine learning methods for some companies which predicted whether some patients would stick to the treatment. This tech helped us understand why some patients drop out and predicted which patients might be at risk of non-adherence in the future. What’s more, it can also keep patients on board by sending them reminders on apps on their phones.’

Deep learning is helping us represent complex information in actionable ways – and we are only just beginning to feel the impact of this technology.

4. Organising huge amounts of data

One of the big promises of AI, and especially deep learning, is the ability to integrate data from different sources into a coherent set.

‘AI will bring great value,’ Silvia comments, ‘when it comes to integrating what can be defined as different modalities. It’s very hard for an individual clinical trial manager to put together information about patients that combines data from different medical fields – from imagery to ‘omics’ (e.g. genomics, metabolomics and proteomics). Yet we need to take all of these into account and draw value out of this big blob of heterogeneous information.

Anyone who has used ChatGPT has witnessed generative AI’s ability to sift through information and organise it clearly at record speed. Having this information at their fingertips will mean doctors can make decisions faster for their patients.

5. Helping doctors do their jobs better

AI-assisted doctors have been a hot topic over the last few years, as they reveal a shift in the way healthcare is delivered to patients.

Nora gives examples of tools already available: ‘LLMs can dig through the clinical records of patients and summarise data for the care team. They can also help doctors by drafting replies to patients’ questions. There are tools that help doctors recommend treatments based on the literature, and others that take notes and queue suggested orders for clinician sign-off. When used with proper human oversight, these tools save time without replacing clinical judgement.’

This has prompted some to wonder about whether we’ll see a drop in quality as these tools are deployed. Will there be a shift away from expert human judgement towards generic, mediocre machine-generated answers?

However, Silvia explains the opposite is happening: ‘Surveys have been carried out with junior doctors to understand what value LLMs are bringing. They were asked whether they see LLMs as oracles, second opinions or something in between. The majority rated the level of expertise of these LLMs as very senior. So they can certainly act as a decision support tool and give confidence, especially to junior doctors who don’t have much experience.’

That said, Silvia also warns of the need to tread carefully in this area. ‘Ultimately, the human relationship with a patient is so important, and it’s impossible to recreate this with AI tools.’

The most successful AI tools don’t replace the clinician. Rather, they support them, make their lives easier and help them sort through vast amounts of data. In his book Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again, Eric Topol goes one step further. He argues that AI will not lessen human contact, but quite the opposite: ‘The greatest opportunity offered by AI is not reducing errors or workloads, or even curing cancer: it is the opportunity to restore the precious and time-honored connection and trust.’ By speeding up admin tasks, doctors will have longer to fully focus on their patients.

Still, it’s important not to rely excessively on these technologies. ‘Doctors will need to be very careful,’ warns Silvia, ‘not to forsake their own instincts and blindly follow instructions from AI tools. This is especially true for more unusual cases. Machine learning tools are better at generalising based on something that happens frequently than at understanding the exception. They’re very weak at adapting to new situations that aren’t represented in the data they have been trained on. These tools are generalists, not specialists.’

Developing AI solutions together

Although both Silvia and Nora are advocates for AI tools in the life sciences, they are also aware of the challenges and risks that lie ahead. There are difficulties to tackle related to infrastructure and staffing, as Christopher Rudolf mentions in his interview with us this month.

But perhaps the biggest obstacle is that mindsets will need to change. Silvia remarks: ‘In some of our partnerships, it’s taken a long time for stakeholders to adopt an AI tool and use it on patients. 90% of the way to adoption is not about innovative potential, but a shift in mindsets. We need to develop tools in close collaboration with medical staff so that they trust these solutions.

So, to reap the full benefits of AI tools in the life sciences, they shouldn’t be developed in a silo. Close collaboration between tech companies and medical staff is the way to produce truly useful tools that are trusted, and bring value, to doctors and patients alike.